1 /* -*- MODE: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mdd) routines
7 * Copyright (C) 2006 Cluster File Systems, Inc.
8 * Author: Wang Di <wangdi@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
29 # define EXPORT_SYMTAB
31 #define DEBUG_SUBSYSTEM S_MDS
33 #include <linux/module.h>
34 #include <linux/jbd.h>
36 #include <obd_class.h>
37 #include <obd_support.h>
38 #include <lprocfs_status.h>
39 /* fid_be_cpu(), fid_cpu_to_be(). */
40 #include <lustre_fid.h>
42 #include <linux/ldiskfs_fs.h>
43 #include <lustre_mds.h>
44 #include <lustre/lustre_idl.h>
46 #include "mdd_internal.h"
48 static struct lu_object_operations mdd_lu_obj_ops;
50 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
51 struct lu_attr *la, struct lustre_capa *capa)
53 LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n",
54 PFID(mdd_object_fid(obj)));
55 return mdo_attr_get(env, obj, la, capa);
58 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
60 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
62 if (flags & LUSTRE_APPEND_FL)
63 obj->mod_flags |= APPEND_OBJ;
65 if (flags & LUSTRE_IMMUTABLE_FL)
66 obj->mod_flags |= IMMUTE_OBJ;
69 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
73 buf = &mdd_env_info(env)->mti_buf;
79 struct llog_cookie *mdd_max_cookie_get(const struct lu_env *env,
80 struct mdd_device *mdd)
82 struct mdd_thread_info *mti = mdd_env_info(env);
85 max_cookie_size = mdd_lov_cookiesize(env, mdd);
86 if (unlikely(mti->mti_max_cookie_size < max_cookie_size)) {
87 if (mti->mti_max_cookie)
88 OBD_FREE(mti->mti_max_cookie, mti->mti_max_cookie_size);
89 mti->mti_max_cookie = NULL;
90 mti->mti_max_cookie_size = 0;
92 if (unlikely(mti->mti_max_cookie == NULL)) {
93 OBD_ALLOC(mti->mti_max_cookie, max_cookie_size);
94 if (unlikely(mti->mti_max_cookie != NULL))
95 mti->mti_max_cookie_size = max_cookie_size;
97 return mti->mti_max_cookie;
100 struct lov_mds_md *mdd_max_lmm_get(const struct lu_env *env,
101 struct mdd_device *mdd)
103 struct mdd_thread_info *mti = mdd_env_info(env);
106 max_lmm_size = mdd_lov_mdsize(env, mdd);
107 if (unlikely(mti->mti_max_lmm_size < max_lmm_size)) {
108 if (mti->mti_max_lmm)
109 OBD_FREE(mti->mti_max_lmm, mti->mti_max_lmm_size);
110 mti->mti_max_lmm = NULL;
111 mti->mti_max_lmm_size = 0;
113 if (unlikely(mti->mti_max_lmm == NULL)) {
114 OBD_ALLOC(mti->mti_max_lmm, max_lmm_size);
115 if (unlikely(mti->mti_max_lmm != NULL))
116 mti->mti_max_lmm_size = max_lmm_size;
118 return mti->mti_max_lmm;
121 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
122 const void *area, ssize_t len)
126 buf = &mdd_env_info(env)->mti_buf;
127 buf->lb_buf = (void *)area;
132 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
134 struct mdd_thread_info *info;
136 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
137 LASSERT(info != NULL);
141 struct lu_object *mdd_object_alloc(const struct lu_env *env,
142 const struct lu_object_header *hdr,
145 struct mdd_object *mdd_obj;
147 OBD_ALLOC_PTR(mdd_obj);
148 if (mdd_obj != NULL) {
151 o = mdd2lu_obj(mdd_obj);
152 lu_object_init(o, NULL, d);
153 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
154 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
155 mdd_obj->mod_count = 0;
156 o->lo_ops = &mdd_lu_obj_ops;
163 static int mdd_object_init(const struct lu_env *env, struct lu_object *o)
165 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
166 struct lu_object *below;
167 struct lu_device *under;
170 under = &d->mdd_child->dd_lu_dev;
171 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
172 mdd_pdlock_init(lu2mdd_obj(o));
176 lu_object_add(o, below);
180 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
182 if (lu_object_exists(o))
183 return mdd_get_flags(env, lu2mdd_obj(o));
188 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
190 struct mdd_object *mdd = lu2mdd_obj(o);
196 static int mdd_object_print(const struct lu_env *env, void *cookie,
197 lu_printer_t p, const struct lu_object *o)
199 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o);
202 /* orphan handling is here */
203 static void mdd_object_delete(const struct lu_env *env,
206 struct mdd_object *mdd_obj = lu2mdd_obj(o);
207 struct thandle *handle = NULL;
210 if (lu2mdd_dev(o->lo_dev)->mdd_orphans == NULL)
213 if (mdd_obj->mod_flags & ORPHAN_OBJ) {
214 mdd_txn_param_build(env, lu2mdd_dev(o->lo_dev),
215 MDD_TXN_INDEX_DELETE_OP);
216 handle = mdd_trans_start(env, lu2mdd_dev(o->lo_dev));
218 CERROR("Cannot get thandle\n");
220 mdd_write_lock(env, mdd_obj);
221 /* let's remove obj from the orphan list */
222 __mdd_orphan_del(env, mdd_obj, handle);
223 mdd_write_unlock(env, mdd_obj);
224 mdd_trans_stop(env, lu2mdd_dev(o->lo_dev),
230 static struct lu_object_operations mdd_lu_obj_ops = {
231 .loo_object_init = mdd_object_init,
232 .loo_object_start = mdd_object_start,
233 .loo_object_free = mdd_object_free,
234 .loo_object_print = mdd_object_print,
235 .loo_object_delete = mdd_object_delete
238 struct mdd_object *mdd_object_find(const struct lu_env *env,
239 struct mdd_device *d,
240 const struct lu_fid *f)
242 struct lu_object *o, *lo;
243 struct mdd_object *m;
246 o = lu_object_find(env, mdd2lu_dev(d)->ld_site, f);
248 m = (struct mdd_object *)o;
250 lo = lu_object_locate(o->lo_header, mdd2lu_dev(d)->ld_type);
251 /* remote object can't be located and should be put then */
253 lu_object_put(env, o);
259 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
261 struct lu_attr *la = &mdd_env_info(env)->mti_la;
265 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
267 mdd_flags_xlate(obj, la->la_flags);
268 if (S_ISDIR(la->la_mode) && la->la_nlink == 1)
269 obj->mod_flags |= MNLINK_OBJ;
274 /* get only inode attributes */
275 int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj,
281 if (ma->ma_valid & MA_INODE)
284 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
285 mdd_object_capa(env, mdd_obj));
287 ma->ma_valid |= MA_INODE;
291 static int mdd_get_default_md(struct mdd_object *mdd_obj,
292 struct lov_mds_md *lmm, int *size)
294 struct lov_desc *ldesc;
295 struct mdd_device *mdd = mdo2mdd(&mdd_obj->mod_obj);
298 ldesc = &mdd->mdd_obd_dev->u.mds.mds_lov_desc;
299 LASSERT(ldesc != NULL);
304 lmm->lmm_magic = LOV_MAGIC_V1;
305 lmm->lmm_object_gr = LOV_OBJECT_GROUP_DEFAULT;
306 lmm->lmm_pattern = ldesc->ld_pattern;
307 lmm->lmm_stripe_size = ldesc->ld_default_stripe_size;
308 lmm->lmm_stripe_count = ldesc->ld_default_stripe_count;
309 *size = sizeof(struct lov_mds_md);
311 RETURN(sizeof(struct lov_mds_md));
314 /* get lov EA only */
315 static int __mdd_lmm_get(const struct lu_env *env,
316 struct mdd_object *mdd_obj, struct md_attr *ma)
321 if (ma->ma_valid & MA_LOV)
324 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size,
327 if (rc == 0 && (ma->ma_need & MA_LOV_DEF)) {
328 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
333 ma->ma_valid |= MA_LOV;
339 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
345 mdd_read_lock(env, mdd_obj);
346 rc = __mdd_lmm_get(env, mdd_obj, ma);
347 mdd_read_unlock(env, mdd_obj);
352 static int __mdd_lmv_get(const struct lu_env *env,
353 struct mdd_object *mdd_obj, struct md_attr *ma)
358 if (ma->ma_valid & MA_LMV)
361 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
364 ma->ma_valid |= MA_LMV;
370 static int mdd_attr_get_internal(const struct lu_env *env,
371 struct mdd_object *mdd_obj,
377 if (ma->ma_need & MA_INODE)
378 rc = mdd_iattr_get(env, mdd_obj, ma);
380 if (rc == 0 && ma->ma_need & MA_LOV) {
381 if (S_ISREG(mdd_object_type(mdd_obj)) ||
382 S_ISDIR(mdd_object_type(mdd_obj)))
383 rc = __mdd_lmm_get(env, mdd_obj, ma);
385 if (rc == 0 && ma->ma_need & MA_LMV) {
386 if (S_ISDIR(mdd_object_type(mdd_obj)))
387 rc = __mdd_lmv_get(env, mdd_obj, ma);
389 #ifdef CONFIG_FS_POSIX_ACL
390 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
391 if (S_ISDIR(mdd_object_type(mdd_obj)))
392 rc = mdd_acl_def_get(env, mdd_obj, ma);
395 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
400 int mdd_attr_get_internal_locked(const struct lu_env *env,
401 struct mdd_object *mdd_obj, struct md_attr *ma)
404 int needlock = ma->ma_need & (MA_LOV | MA_LMV | MA_ACL_DEF);
407 mdd_read_lock(env, mdd_obj);
408 rc = mdd_attr_get_internal(env, mdd_obj, ma);
410 mdd_read_unlock(env, mdd_obj);
415 * No permission check is needed.
417 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
420 struct mdd_object *mdd_obj = md2mdd_obj(obj);
424 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
429 * No permission check is needed.
431 static int mdd_xattr_get(const struct lu_env *env,
432 struct md_object *obj, struct lu_buf *buf,
435 struct mdd_object *mdd_obj = md2mdd_obj(obj);
440 LASSERT(mdd_object_exists(mdd_obj));
442 mdd_read_lock(env, mdd_obj);
443 rc = mdo_xattr_get(env, mdd_obj, buf, name,
444 mdd_object_capa(env, mdd_obj));
445 mdd_read_unlock(env, mdd_obj);
451 * Permission check is done when open,
452 * no need check again.
454 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
457 struct mdd_object *mdd_obj = md2mdd_obj(obj);
458 struct dt_object *next;
463 LASSERT(mdd_object_exists(mdd_obj));
465 next = mdd_object_child(mdd_obj);
466 mdd_read_lock(env, mdd_obj);
467 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
468 mdd_object_capa(env, mdd_obj));
469 mdd_read_unlock(env, mdd_obj);
474 * No permission check is needed.
476 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
479 struct mdd_object *mdd_obj = md2mdd_obj(obj);
484 mdd_read_lock(env, mdd_obj);
485 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
486 mdd_read_unlock(env, mdd_obj);
491 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
492 struct mdd_object *c, struct md_attr *ma,
493 struct thandle *handle)
495 struct lu_attr *attr = &ma->ma_attr;
496 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
500 if (!mdd_object_exists(c)) {
501 struct dt_object *next = mdd_object_child(c);
504 /* @hint will be initialized by underlying device. */
505 next->do_ops->do_ah_init(env, hint,
506 p ? mdd_object_child(p) : NULL,
507 attr->la_mode & S_IFMT);
508 rc = mdo_create_obj(env, c, attr, hint, handle);
509 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
516 static inline int mdd_attr_check(const struct lu_env *env,
517 struct mdd_object *obj,
518 struct lu_attr *attr)
520 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
524 if (attr->la_valid & LA_CTIME) {
525 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
529 if (attr->la_ctime < tmp_la->la_ctime)
530 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
531 else if (attr->la_valid == LA_CTIME &&
532 attr->la_ctime == tmp_la->la_ctime)
533 attr->la_valid &= ~LA_CTIME;
538 int mdd_attr_set_internal(const struct lu_env *env,
539 struct mdd_object *obj,
540 struct lu_attr *attr,
541 struct thandle *handle,
547 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
548 #ifdef CONFIG_FS_POSIX_ACL
549 if (!rc && (attr->la_valid & LA_MODE) && needacl)
550 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
555 int mdd_attr_check_set_internal(const struct lu_env *env,
556 struct mdd_object *obj,
557 struct lu_attr *attr,
558 struct thandle *handle,
564 rc = mdd_attr_check(env, obj, attr);
569 rc = mdd_attr_set_internal(env, obj, attr, handle, needacl);
573 static int mdd_attr_set_internal_locked(const struct lu_env *env,
574 struct mdd_object *obj,
575 struct lu_attr *attr,
576 struct thandle *handle,
582 needacl = needacl && (attr->la_valid & LA_MODE);
585 mdd_write_lock(env, obj);
587 rc = mdd_attr_set_internal(env, obj, attr, handle, needacl);
590 mdd_write_unlock(env, obj);
594 int mdd_attr_check_set_internal_locked(const struct lu_env *env,
595 struct mdd_object *obj,
596 struct lu_attr *attr,
597 struct thandle *handle,
603 rc = mdd_attr_check(env, obj, attr);
608 rc = mdd_attr_set_internal_locked(env, obj, attr, handle,
613 static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj,
614 const struct lu_buf *buf, const char *name,
615 int fl, struct thandle *handle)
617 struct lustre_capa *capa = mdd_object_capa(env, obj);
621 if (buf->lb_buf && buf->lb_len > 0)
622 rc = mdo_xattr_set(env, obj, buf, name, 0, handle, capa);
623 else if (buf->lb_buf == NULL && buf->lb_len == 0)
624 rc = mdo_xattr_del(env, obj, name, handle, capa);
630 * This gives the same functionality as the code between
631 * sys_chmod and inode_setattr
632 * chown_common and inode_setattr
633 * utimes and inode_setattr
634 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
636 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
637 struct lu_attr *la, const struct md_attr *ma)
639 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
640 struct md_ucred *uc = md_ucred(env);
647 /* Do not permit change file type */
648 if (la->la_valid & LA_TYPE)
651 /* They should not be processed by setattr */
652 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
655 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
659 /* This is only for set ctime when rename's source is on remote MDS. */
660 if (unlikely(la->la_valid == LA_CTIME)) {
661 rc = mdd_may_delete(env, NULL, obj, (struct md_attr *)ma, 1, 0);
662 if (rc == 0 && la->la_ctime <= tmp_la->la_ctime)
663 la->la_valid &= ~LA_CTIME;
667 if (la->la_valid == LA_ATIME) {
668 /* This is atime only set for read atime update on close. */
669 if (la->la_atime <= tmp_la->la_atime +
670 mdd_obj2mdd_dev(obj)->mdd_atime_diff)
671 la->la_valid &= ~LA_ATIME;
675 /* Check if flags change. */
676 if (la->la_valid & LA_FLAGS) {
677 unsigned int oldflags = 0;
678 unsigned int newflags = la->la_flags &
679 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
681 if ((uc->mu_fsuid != tmp_la->la_uid) &&
682 !mdd_capable(uc, CAP_FOWNER))
685 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
686 * only be changed by the relevant capability. */
687 if (mdd_is_immutable(obj))
688 oldflags |= LUSTRE_IMMUTABLE_FL;
689 if (mdd_is_append(obj))
690 oldflags |= LUSTRE_APPEND_FL;
691 if ((oldflags ^ newflags) &&
692 !mdd_capable(uc, CAP_LINUX_IMMUTABLE))
695 if (!S_ISDIR(tmp_la->la_mode))
696 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
699 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
700 (la->la_valid & ~LA_FLAGS) &&
701 !(ma->ma_attr_flags & MDS_PERM_BYPASS))
704 /* Check for setting the obj time. */
705 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
706 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
707 if ((uc->mu_fsuid != tmp_la->la_uid) &&
708 !mdd_capable(uc, CAP_FOWNER)) {
709 rc = mdd_permission_internal_locked(env, obj, tmp_la,
716 /* Make sure a caller can chmod. */
717 if (la->la_valid & LA_MODE) {
719 * Bypass la_vaild == LA_MODE,
720 * this is for changing file with SUID or SGID.
722 if ((la->la_valid & ~LA_MODE) &&
723 (uc->mu_fsuid != tmp_la->la_uid) &&
724 !mdd_capable(uc, CAP_FOWNER))
727 if (la->la_mode == (umode_t) -1)
728 la->la_mode = tmp_la->la_mode;
730 la->la_mode = (la->la_mode & S_IALLUGO) |
731 (tmp_la->la_mode & ~S_IALLUGO);
733 /* Also check the setgid bit! */
734 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ? la->la_gid :
735 tmp_la->la_gid) && !mdd_capable(uc, CAP_FSETID))
736 la->la_mode &= ~S_ISGID;
738 la->la_mode = tmp_la->la_mode;
741 /* Make sure a caller can chown. */
742 if (la->la_valid & LA_UID) {
743 if (la->la_uid == (uid_t) -1)
744 la->la_uid = tmp_la->la_uid;
745 if (((uc->mu_fsuid != tmp_la->la_uid) ||
746 (la->la_uid != tmp_la->la_uid)) &&
747 !mdd_capable(uc, CAP_CHOWN))
751 * If the user or group of a non-directory has been
752 * changed by a non-root user, remove the setuid bit.
753 * 19981026 David C Niemi <niemi@tux.org>
755 * Changed this to apply to all users, including root,
756 * to avoid some races. This is the behavior we had in
757 * 2.0. The check for non-root was definitely wrong
758 * for 2.2 anyway, as it should have been using
759 * CAP_FSETID rather than fsuid -- 19990830 SD.
761 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
762 !S_ISDIR(tmp_la->la_mode)) {
763 la->la_mode &= ~S_ISUID;
764 la->la_valid |= LA_MODE;
768 /* Make sure caller can chgrp. */
769 if (la->la_valid & LA_GID) {
770 if (la->la_gid == (gid_t) -1)
771 la->la_gid = tmp_la->la_gid;
772 if (((uc->mu_fsuid != tmp_la->la_uid) ||
773 ((la->la_gid != tmp_la->la_gid) &&
774 !lustre_in_group_p(uc, la->la_gid))) &&
775 !mdd_capable(uc, CAP_CHOWN))
779 * Likewise, if the user or group of a non-directory
780 * has been changed by a non-root user, remove the
781 * setgid bit UNLESS there is no group execute bit
782 * (this would be a file marked for mandatory
783 * locking). 19981026 David C Niemi <niemi@tux.org>
785 * Removed the fsuid check (see the comment above) --
788 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
789 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
790 la->la_mode &= ~S_ISGID;
791 la->la_valid |= LA_MODE;
795 if (la->la_valid & (LA_SIZE | LA_BLOCKS) &&
796 !(la->la_valid & LA_TRUNC)) {
797 /* For "Size-on-MDS" case, the MAY_WRITE perm
798 * has been checked when file open. */
800 /* For the "Size-on-MDS" setattr update, merge coming
801 * attributes with the set in the inode. BUG 10641 */
802 if ((la->la_valid & LA_ATIME) &&
803 (la->la_atime <= tmp_la->la_atime))
804 la->la_valid &= ~LA_ATIME;
806 /* OST attributes do not have a priority over MDS attributes,
807 * so drop times if ctime is equal. */
808 if ((la->la_valid & LA_CTIME) &&
809 (la->la_ctime <= tmp_la->la_ctime))
810 la->la_valid &= ~(LA_MTIME | LA_CTIME);
812 if (la->la_valid & LA_TRUNC) {
813 /* For truncate, we should have MAY_WRITE perm. */
814 if (!((la->la_valid & MDS_OPEN_OWNEROVERRIDE) &&
815 (uc->mu_fsuid == tmp_la->la_uid)) &&
816 !(ma->ma_attr_flags & MDS_PERM_BYPASS)) {
817 rc = mdd_permission_internal_locked(env, obj,
823 if (la->la_valid & LA_CTIME) {
824 /* The pure setattr, it has the priority over what is
825 * already set, do not drop it if ctime is equal. */
826 if (la->la_ctime < tmp_la->la_ctime)
827 la->la_valid &= ~(LA_ATIME | LA_MTIME |
835 /* set attr and LOV EA at once, return updated attr */
836 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
837 const struct md_attr *ma)
839 struct mdd_object *mdd_obj = md2mdd_obj(obj);
840 struct mdd_device *mdd = mdo2mdd(obj);
841 struct thandle *handle;
842 struct lov_mds_md *lmm = NULL;
843 struct llog_cookie *logcookies = NULL;
844 int rc, lmm_size = 0, cookie_size = 0;
845 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
848 mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP);
849 handle = mdd_trans_start(env, mdd);
851 RETURN(PTR_ERR(handle));
852 /*TODO: add lock here*/
853 /* start a log jounal handle if needed */
854 if (S_ISREG(mdd_object_type(mdd_obj)) &&
855 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
856 lmm_size = mdd_lov_mdsize(env, mdd);
857 lmm = mdd_max_lmm_get(env, mdd);
859 GOTO(cleanup, rc = -ENOMEM);
861 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
868 if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME))
869 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
870 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
872 *la_copy = ma->ma_attr;
873 rc = mdd_fix_attr(env, mdd_obj, la_copy, ma);
877 if (la_copy->la_valid & LA_FLAGS) {
878 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
881 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
882 } else if (la_copy->la_valid) { /* setattr */
883 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
885 /* journal chown/chgrp in llog, just like unlink */
886 if (rc == 0 && lmm_size){
887 cookie_size = mdd_lov_cookiesize(env, mdd);
888 logcookies = mdd_max_cookie_get(env, mdd);
889 if (logcookies == NULL)
890 GOTO(cleanup, rc = -ENOMEM);
892 if (mdd_setattr_log(env, mdd, ma, lmm, lmm_size,
893 logcookies, cookie_size) <= 0)
898 if (rc == 0 && ma->ma_valid & MA_LOV) {
901 mode = mdd_object_type(mdd_obj);
902 if (S_ISREG(mode) || S_ISDIR(mode)) {
903 rc = mdd_lsm_sanity_check(env, mdd_obj);
907 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
908 ma->ma_lmm_size, handle, 1);
913 mdd_trans_stop(env, mdd, rc, handle);
914 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
915 /*set obd attr, if needed*/
916 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size,
922 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
923 const struct lu_buf *buf, const char *name, int fl,
924 struct thandle *handle)
929 mdd_write_lock(env, obj);
930 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
931 mdd_write_unlock(env, obj);
936 static int mdd_xattr_sanity_check(const struct lu_env *env,
937 struct mdd_object *obj)
939 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
940 struct md_ucred *uc = md_ucred(env);
944 if (mdd_is_immutable(obj) || mdd_is_append(obj))
947 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
951 if ((uc->mu_fsuid != tmp_la->la_uid) && !mdd_capable(uc, CAP_FOWNER))
957 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
958 const struct lu_buf *buf, const char *name,
959 int fl, const struct lu_attr *la)
961 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
962 struct mdd_object *mdd_obj = md2mdd_obj(obj);
963 struct mdd_device *mdd = mdo2mdd(obj);
964 struct thandle *handle;
968 rc = mdd_xattr_sanity_check(env, mdd_obj);
972 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
973 handle = mdd_trans_start(env, mdd);
975 RETURN(PTR_ERR(handle));
977 rc = mdd_xattr_set_txn(env, mdd_obj, buf, name, fl, handle);
978 if (rc == 0 && likely(la != NULL)) {
979 LASSERT(la->la_valid & LA_CTIME);
980 la_copy->la_ctime = la->la_ctime;
981 la_copy->la_valid = LA_CTIME;
982 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la_copy,
985 mdd_trans_stop(env, mdd, rc, handle);
990 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
991 const char *name, const struct lu_attr *la)
993 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
994 struct mdd_object *mdd_obj = md2mdd_obj(obj);
995 struct mdd_device *mdd = mdo2mdd(obj);
996 struct thandle *handle;
1000 rc = mdd_xattr_sanity_check(env, mdd_obj);
1004 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1005 handle = mdd_trans_start(env, mdd);
1007 RETURN(PTR_ERR(handle));
1009 mdd_write_lock(env, mdd_obj);
1010 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1011 mdd_object_capa(env, mdd_obj));
1012 mdd_write_unlock(env, mdd_obj);
1013 if (rc == 0 && likely(la != NULL)) {
1014 LASSERT(la->la_valid & LA_CTIME);
1015 la_copy->la_ctime = la->la_ctime;
1016 la_copy->la_valid = LA_CTIME;
1017 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la_copy,
1021 mdd_trans_stop(env, mdd, rc, handle);
1026 /* partial unlink */
1027 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
1030 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1031 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1032 struct mdd_device *mdd = mdo2mdd(obj);
1033 struct thandle *handle;
1038 * Check -ENOENT early here because we need to get object type
1039 * to calculate credits before transaction start
1041 if (!mdd_object_exists(mdd_obj))
1044 LASSERT(mdd_object_exists(mdd_obj) > 0);
1046 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
1050 handle = mdd_trans_start(env, mdd);
1054 mdd_write_lock(env, mdd_obj);
1056 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
1060 __mdd_ref_del(env, mdd_obj, handle, 0);
1062 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
1064 __mdd_ref_del(env, mdd_obj, handle, 1);
1067 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1068 la_copy->la_ctime = ma->ma_attr.la_ctime;
1070 la_copy->la_valid = LA_CTIME;
1071 rc = mdd_attr_check_set_internal(env, mdd_obj, la_copy, handle, 0);
1075 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
1079 mdd_write_unlock(env, mdd_obj);
1080 mdd_trans_stop(env, mdd, rc, handle);
1084 /* partial operation */
1085 static int mdd_oc_sanity_check(const struct lu_env *env,
1086 struct mdd_object *obj,
1092 switch (ma->ma_attr.la_mode & S_IFMT) {
1109 static int mdd_object_create(const struct lu_env *env,
1110 struct md_object *obj,
1111 const struct md_op_spec *spec,
1115 struct mdd_device *mdd = mdo2mdd(obj);
1116 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1117 const struct lu_fid *pfid = spec->u.sp_pfid;
1118 struct thandle *handle;
1122 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
1123 handle = mdd_trans_start(env, mdd);
1125 RETURN(PTR_ERR(handle));
1127 mdd_write_lock(env, mdd_obj);
1128 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
1132 rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle);
1136 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1137 /* If creating the slave object, set slave EA here. */
1138 int lmv_size = spec->u.sp_ea.eadatalen;
1139 struct lmv_stripe_md *lmv;
1141 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
1142 LASSERT(lmv != NULL && lmv_size > 0);
1144 rc = __mdd_xattr_set(env, mdd_obj,
1145 mdd_buf_get_const(env, lmv, lmv_size),
1146 MDS_LMV_MD_NAME, 0, handle);
1150 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr,
1153 #ifdef CONFIG_FS_POSIX_ACL
1154 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
1155 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1157 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
1158 buf->lb_len = spec->u.sp_ea.eadatalen;
1159 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1160 rc = __mdd_acl_init(env, mdd_obj, buf,
1161 &ma->ma_attr.la_mode,
1166 ma->ma_attr.la_valid |= LA_MODE;
1169 pfid = spec->u.sp_ea.fid;
1172 rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle);
1176 mdd_write_unlock(env, mdd_obj);
1178 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
1180 mdd_trans_stop(env, mdd, rc, handle);
1185 static int mdd_ref_add(const struct lu_env *env, struct md_object *obj,
1186 const struct md_attr *ma)
1188 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1189 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1190 struct mdd_device *mdd = mdo2mdd(obj);
1191 struct thandle *handle;
1195 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1196 handle = mdd_trans_start(env, mdd);
1200 mdd_write_lock(env, mdd_obj);
1201 rc = mdd_link_sanity_check(env, NULL, NULL, mdd_obj);
1203 __mdd_ref_add(env, mdd_obj, handle);
1204 mdd_write_unlock(env, mdd_obj);
1206 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1207 la_copy->la_ctime = ma->ma_attr.la_ctime;
1209 la_copy->la_valid = LA_CTIME;
1210 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la_copy,
1213 mdd_trans_stop(env, mdd, 0, handle);
1219 * do NOT or the MAY_*'s, you'll get the weakest
1221 int accmode(const struct lu_env *env, struct lu_attr *la, int flags)
1225 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1226 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1227 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1228 * owner can write to a file even if it is marked readonly to hide
1229 * its brokenness. (bug 5781) */
1230 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1231 struct md_ucred *uc = md_ucred(env);
1233 if ((uc == NULL) || (uc->mu_valid == UCRED_INIT) ||
1234 (la->la_uid == uc->mu_fsuid))
1238 if (flags & FMODE_READ)
1240 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1242 if (flags & MDS_FMODE_EXEC)
1247 static int mdd_open_sanity_check(const struct lu_env *env,
1248 struct mdd_object *obj, int flag)
1250 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1255 if (mdd_is_dead_obj(obj))
1258 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1262 if (S_ISLNK(tmp_la->la_mode))
1265 mode = accmode(env, tmp_la, flag);
1267 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1270 if (!(flag & MDS_OPEN_CREATED)) {
1271 rc = mdd_permission_internal(env, obj, tmp_la, mode);
1276 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1277 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1278 flag &= ~MDS_OPEN_TRUNC;
1280 /* For writing append-only file must open it with append mode. */
1281 if (mdd_is_append(obj)) {
1282 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1284 if (flag & MDS_OPEN_TRUNC)
1290 * Now, flag -- O_NOATIME does not be packed by client.
1292 if (flag & O_NOATIME) {
1293 struct md_ucred *uc = md_ucred(env);
1295 if (uc && ((uc->mu_valid == UCRED_OLD) ||
1296 (uc->mu_valid == UCRED_NEW)) &&
1297 (uc->mu_fsuid != tmp_la->la_uid) &&
1298 !mdd_capable(uc, CAP_FOWNER))
1306 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1309 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1312 mdd_write_lock(env, mdd_obj);
1314 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1316 mdd_obj->mod_count++;
1318 mdd_write_unlock(env, mdd_obj);
1322 /* return md_attr back,
1323 * if it is last unlink then return lov ea + llog cookie*/
1324 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1330 if (S_ISREG(mdd_object_type(obj))) {
1331 /* Return LOV & COOKIES unconditionally here. We clean evth up.
1332 * Caller must be ready for that. */
1333 rc = __mdd_lmm_get(env, obj, ma);
1334 if ((ma->ma_valid & MA_LOV))
1335 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
1342 * No permission check is needed.
1344 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1348 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1349 struct thandle *handle;
1352 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
1355 handle = mdd_trans_start(env, mdo2mdd(obj));
1357 RETURN(PTR_ERR(handle));
1359 mdd_write_lock(env, mdd_obj);
1360 /* release open count */
1361 mdd_obj->mod_count --;
1363 rc = mdd_iattr_get(env, mdd_obj, ma);
1364 if (rc == 0 && mdd_obj->mod_count == 0 && ma->ma_attr.la_nlink == 0)
1365 rc = mdd_object_kill(env, mdd_obj, ma);
1367 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
1369 mdd_write_unlock(env, mdd_obj);
1370 mdd_trans_stop(env, mdo2mdd(obj), rc, handle);
1375 * Permission check is done when open,
1376 * no need check again.
1378 static int mdd_readpage_sanity_check(const struct lu_env *env,
1379 struct mdd_object *obj)
1381 struct dt_object *next = mdd_object_child(obj);
1385 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1393 static int mdd_dir_page_build(const struct lu_env *env, int first,
1394 void *area, int nob, struct dt_it_ops *iops,
1395 struct dt_it *it, __u64 *start, __u64 *end,
1396 struct lu_dirent **last)
1398 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
1399 struct mdd_thread_info *info = mdd_env_info(env);
1400 struct lu_fid_pack *pack = &info->mti_pack;
1402 struct lu_dirent *ent;
1405 memset(area, 0, sizeof (struct lu_dirpage));
1406 area += sizeof (struct lu_dirpage);
1407 nob -= sizeof (struct lu_dirpage);
1410 LASSERT(nob > sizeof *ent);
1420 name = (char *)iops->key(env, it);
1421 len = iops->key_size(env, it);
1423 pack = (struct lu_fid_pack *)iops->rec(env, it);
1424 result = fid_unpack(pack, fid);
1428 recsize = (sizeof(*ent) + len + 7) & ~7;
1429 hash = iops->store(env, it);
1432 CDEBUG(D_INFO, "%p %p %d "DFID": "LPU64" (%d) \"%*.*s\"\n",
1433 name, ent, nob, PFID(fid), hash, len, len, len, name);
1435 if (nob >= recsize) {
1436 ent->lde_fid = *fid;
1437 fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid);
1438 ent->lde_hash = hash;
1439 ent->lde_namelen = cpu_to_le16(len);
1440 ent->lde_reclen = cpu_to_le16(recsize);
1441 memcpy(ent->lde_name, name, len);
1442 if (first && ent == area)
1445 ent = (void *)ent + recsize;
1447 result = iops->next(env, it);
1450 * record doesn't fit into page, enlarge previous one.
1452 LASSERT(*last != NULL);
1453 (*last)->lde_reclen =
1454 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
1458 } while (result == 0);
1463 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
1464 const struct lu_rdpg *rdpg)
1467 struct dt_object *next = mdd_object_child(obj);
1468 struct dt_it_ops *iops;
1470 struct lu_dirent *last;
1477 LASSERT(rdpg->rp_pages != NULL);
1478 LASSERT(next->do_index_ops != NULL);
1480 if (rdpg->rp_count <= 0)
1484 * iterate through directory and fill pages from @rdpg
1486 iops = &next->do_index_ops->dio_it;
1487 it = iops->init(env, next, 0, mdd_object_capa(env, obj));
1491 rc = iops->load(env, it, rdpg->rp_hash);
1495 * Iterator didn't find record with exactly the key requested.
1497 * It is currently either
1499 * - positioned above record with key less than
1500 * requested---skip it.
1502 * - or not positioned at all (is in IAM_IT_SKEWED
1503 * state)---position it on the next item.
1505 rc = iops->next(env, it);
1510 * At this point and across for-loop:
1512 * rc == 0 -> ok, proceed.
1513 * rc > 0 -> end of directory.
1516 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
1517 i++, nob -= CFS_PAGE_SIZE) {
1518 LASSERT(i < rdpg->rp_npages);
1519 pg = rdpg->rp_pages[i];
1520 rc = mdd_dir_page_build(env, !i, cfs_kmap(pg),
1521 min_t(int, nob, CFS_PAGE_SIZE), iops,
1522 it, &hash_start, &hash_end, &last);
1523 if (rc != 0 || i == rdpg->rp_npages - 1)
1524 last->lde_reclen = 0;
1531 hash_end = DIR_END_OFF;
1535 struct lu_dirpage *dp;
1537 dp = cfs_kmap(rdpg->rp_pages[0]);
1538 dp->ldp_hash_start = rdpg->rp_hash;
1539 dp->ldp_hash_end = hash_end;
1542 * No pages were processed, mark this.
1544 dp->ldp_flags |= LDF_EMPTY;
1545 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
1546 cfs_kunmap(rdpg->rp_pages[0]);
1549 iops->fini(env, it);
1554 static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1555 const struct lu_rdpg *rdpg)
1557 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1561 LASSERT(mdd_object_exists(mdd_obj));
1563 mdd_read_lock(env, mdd_obj);
1564 rc = mdd_readpage_sanity_check(env, mdd_obj);
1566 GOTO(out_unlock, rc);
1568 if (mdd_is_dead_obj(mdd_obj)) {
1570 struct lu_dirpage *dp;
1573 * According to POSIX, please do not return any entry to client:
1574 * even dot and dotdot should not be returned.
1576 CWARN("readdir from dead object: "DFID"\n",
1577 PFID(mdd_object_fid(mdd_obj)));
1579 if (rdpg->rp_count <= 0)
1580 GOTO(out_unlock, rc = -EFAULT);
1581 LASSERT(rdpg->rp_pages != NULL);
1583 pg = rdpg->rp_pages[0];
1584 dp = (struct lu_dirpage*)cfs_kmap(pg);
1585 memset(dp, 0 , sizeof(struct lu_dirpage));
1586 dp->ldp_hash_start = rdpg->rp_hash;
1587 dp->ldp_hash_end = DIR_END_OFF;
1588 dp->ldp_flags |= LDF_EMPTY;
1589 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
1591 GOTO(out_unlock, rc = 0);
1594 rc = __mdd_readpage(env, mdd_obj, rdpg);
1598 mdd_read_unlock(env, mdd_obj);
1602 struct md_object_operations mdd_obj_ops = {
1603 .moo_permission = mdd_permission,
1604 .moo_attr_get = mdd_attr_get,
1605 .moo_attr_set = mdd_attr_set,
1606 .moo_xattr_get = mdd_xattr_get,
1607 .moo_xattr_set = mdd_xattr_set,
1608 .moo_xattr_list = mdd_xattr_list,
1609 .moo_xattr_del = mdd_xattr_del,
1610 .moo_object_create = mdd_object_create,
1611 .moo_ref_add = mdd_ref_add,
1612 .moo_ref_del = mdd_ref_del,
1613 .moo_open = mdd_open,
1614 .moo_close = mdd_close,
1615 .moo_readpage = mdd_readpage,
1616 .moo_readlink = mdd_readlink,
1617 .moo_capa_get = mdd_capa_get